This thesis investigates the analysis and optimization of safety and performance-critical computational tasks for autonomous robots, operating in unknown and unstructured environments with complex objectives under strict computational and power constraints. Our primary contribution is a novel safety-performance (SP) metric that emphasizes on safety while rewarding enhanced performance of real-time computational tasks, expanding the notion of nominal safety in the autonomous vehicle domain. We adopt the Stochastic Heterogeneous Parallel Directed Acyclic Graph (SHP-DAG) model to capture the uncertain nature of robotic applications and their required computations, modeling execution times using probability distributions instead of deterministic worst-case execution time (WCET). We argue that computational tasks enabling robotic autonomy, such as localization and mapping, path planning, task allocation, depth estimation, and optical flow, must be scheduled and optimized to guarantee timely and correct behavior while allowing for runtime reconfiguration of scheduling parameters. To attain computational awareness in autonomous robots, we conduct a data-driven study of these computational tasks from the resource management perspective, profiling and analyzing their timing, power, and memory performance across three embedded computing platforms. Our SP metric allows us to apply the schedulers First-In-First-Out (FIFO) and Completely Fair Scheduler (CFS) of the Linux kernel on complex robotic computational tasks and compare the SP metric with baseline metrics, such as average and worst-case makespan. Extensive experimental results on NVIDIA Jetson AGX Xavier hardware demonstrate the effectiveness of the proposed SP metric in managing computational tasks while balancing safety and performance in robotic systems. Our findings reveal a correlation between task performance and a robot's operational environment, which justifies the concept of computation-aware robots and highlights the importance of our work as a crucial step towards this goal. Finally, we also integrate a custom scheduler with the FIFO priorities with our SHP-DAG and show the efficacy of our framework in comparison to default fair scheduler. / Doctor of Philosophy / This paper explores how to improve the safety and performance of autonomous robots operating in unpredictable and complex environments. These robots need to carry out various tasks such as mapping, path planning, and depth estimation, while managing limited computing power and energy resources. To achieve this, we introduce a new safety-performance (SP) metric that prioritizes safety while rewarding better task performance. We use a cutting-edge model that captures the uncertainty of robotic tasks and their required computing resources. By doing so, we can better schedule and optimize these tasks to ensure timely and correct behavior while allowing for adjustments to scheduling parameters during operation. Our study investigates the performance of key computing tasks on various embedded computing platforms. By comparing our SP metric with traditional measures, we can demonstrate the effectiveness of our approach in managing these tasks while balancing safety and performance in robotic systems. We also do system integration of a real-time scheduler with robotic tasks, which shows the efficacy of our framework. Our findings show a connection between a robot's environment and its computing performance, highlighting the importance of our work as a critical step towards creating smarter and safer autonomous robots that can better adapt to their surroundings.

This thesis examines a manufacturing process using a real time interface with a 6502 microprocessor that gives the appearance of parallel processing. Two separate processes are operated, apparently simultaneously with an asynchronous interface between the two processes. An Apple microcomputer, an ISAAC data transfer system and a constructed simulation model are used to demonstrate this process. The model is constructed of Fischertechnik manufactured parts for the support framework, as well as gearing devices, small DC electric motors, and sensing devices in the form of photo-electric switches and single pole double throw switches physically activated by the constructed model. The software, written in Applesoft BASIC and Cyborg's Labsoft, was designed to operate the modeled processes simultaneously and allow an asynchronous interaction between the two processes. The model has applications for use as a method to illustrate manufacturing techniques and to assist in the design and control of manufacturing processes.

Deregulation of electricity markets, increased usage of intermittent energy sources, and growing environmental concerns have created a volatile process manufacturing environment. Survival under this new paradigm requires chemical manufactures to shift from the traditional steady-state operation to a more dynamic and flexible operation mode. Under more frequent operating changes, the transition dynamics become increasingly relevant, rendering the traditional steady-state based scheduling decision-making suboptimal. This has motivated calls for the integration of scheduling and control. In an integrated scheduling and control framework, the scheduling decisions are based on a dynamic representation of the process. While various integration paradigms are proposed in the literature, our study concentrates on the closed-loop integration of scheduling and control. There are two main advantages to this approach: (i) seamless integration with the existing control system (i.e. it does not require a new control system infrastructure), (ii) the framework is aware of the control system dynamics, and hence has knowledge of the closed-loop process dynamics. The later aspect is particularly important as the control system plays a key role in determining the transition dynamics. The first part of our work is dedicated to developing an integrated scheduling and control framework that computes set-point trajectories, to be tracked by the lower-level linear model predictive control system, that are robust to demand uncertainty. We employ a piecewise linear representation of the nonlinear process model to obtain a mixed-integer linear programming (MILP) problem, thus alleviating the computational complexity compared to a mixed-integer nonlinear programming formulation. The value of the stochastic solution is used to confirm the superiority of the robust formulation against a nominal one that disregards uncertainty. In the second part of this study, we expand the framework to accommodate additional uncertainty types, including model and cost uncertainty. In the third part of this thesis, a deterministic integrated scheduling and control framework for processes controlled by distributed linear model predictive control is developed. The integrated problem is formulated as a MILP. To reduce the solution time, we introduce strategies to approximate the feedback control action. Through case studies, we demonstrate that knowledge of the control system enables the framework to effectively coordinate the MPC subsystems. The framework performs well even under conditions of plant-model mismatch conditions. In the final part of this study, we introduce an integrated scheduling and control formulation for processes controlled by nonlinear model predictive control (NMPC). Here, discrete scheduling decisions are represented using complementarity conditions. Additionally, we use the first-order Karush-Kuhn-Tucker conditions of the NMPC controller to compute the input values in the integrated problem. The resulting problem is a mathematical program with complementarity constraints that we solve using a regularization approach. For all case studies, the complementarity formulation effectively capture discrete scheduling decisions, and the KKT conditions always provides a local optimum of the associated NMPC problem. In summary, this study of the integration of scheduling and control addresses various control systems, uncertainty, and strategies for enhancing the solution time. Furthermore, we assess the performance of the proposed frameworks under conditions of plant-model mismatch, a common scenario in real-life applications. / Thesis / Doctor of Philosophy (PhD)

In real-word manufacturing environments, finding the right job sequences and their associated schedules when resource, precedence, and timing constraints are imposed is a difficult task. For most practical problems classical scheduling easily leads to an exponential growth in the number of possible schedules. Moreover, a decision time period of hours or even minutes is too long. Good solutions are often needed in real-time. The problem becomes even more complicated if changes, such as new orders or resource breakdowns, occur within the manufacturing system. One approach to overcome the challenges of solving classical scheduling problems is the use of distributed schemes such as agent or holonic-based control architectures. This dissertation presents an innovative control architecture that uses the holonic concept, capable of delivering good solutions when applied in dynamic environments. The general holonic control framework presented in this research has specific characteristics not found in others reported so far. Using a modular approach it takes into account all the categories of hardware and software resources of a manufacturing system. Due to its modularity, the holonic control framework can be used for assigning and scheduling different task types, separately or simultaneously. Thus, it can be used not only for assigning and scheduling transport tasks, but also for finding feasible solutions to the job assignment and scheduling of processing tasks, or to better utilize the auxiliary equipment and devices in a manufacturing system. In the holonic system, under real-time constraints, a feasible schedule for the material handling resources emerges from the combination of individual holon's schedules. Internal evaluation algorithms and coordination mechanisms between the entities in the architecture form the basis for the resultant schedules. The experimental results obtained show a percentage difference between the makespan values obtained using the holonic scheduling approach and the optimal values of under seven percent. Since current control systems in use in industry lack the ability to adapt to dynamic manufacturing environments, the holonic architecture designed and the tests performed in this research could be a part in the effort to build the foundations for the control systems of the next generation manufacturing systems. / Ph. D.

Utility Accrual (UA) scheduling is a method of dynamic real-time scheduling that is designed to respond to overload conditions by producing a feasible schedule that heuristically maximizes a pre-defined metric of utility. Whereas utility accrual schedulers have traditionally focused on CPU overload, this dissertation explores memory overload conditions during which the aggregate memory demand exceeds a system's available memory bandwidth. Real-time systems are typically implemented in C or other languages that use explicit dynamic memory management. Taking advantage of modern type-safe languages, such as Java, necessitates the use of garbage collection (GC). The timeliness requirements of real-time systems, however, impose specific demands on the garbage collector. Garbage collection introduces a significant source of unpredictability in the execution timeline of a task because it unexpectedly interjects pauses of arbitrary length, at arbitrary points in time, with an arbitrary frequency. To construct a feasible schedule, a real-time scheduler must have the ability to predict the collector's activities and plan for them accordingly. We have devised CADUS (Collector-Aware Dynamic Utility Scheduler), a utility accrual algorithm that tightly links CPU scheduling with the memory requirements -and the corresponding garbage collection activities - of real-time tasks. By constructing and storing memory time allocation profiles, we address the problem of GC activation strategy. We estimate GC latency by using a real-time collector and modeling its behavior. We project GC frequency by planning, at schedule construction time, the memory bandwidth available to the collector. CADUS can point the collector's activities to any specific task in the system. The runtime system provides this ability by maintaining separate logical heaps for all tasks. We demonstrate the viability of CADUS through extensive simulation studies. We evaluated the behavior of CADUS under a wide range of CPU and memory load conditions and utility distributions. We compared its performance against an existing GC-unaware UA scheduler and found that CADUS consistently outperformed its GC-unaware counterpart. We investigated and identified the reasons for the superior performance of CADUS and quantified our results. Most significantly, we found that in an overloaded dynamic soft real-time system, a scheduler's preemption decisions have a highly significant impact on GC latency. A dynamic real-time scheduler therefore must predict the impact of its preemption decisions on GC latency in order to construct time-feasible schedules. / Ph. D.

Due to close coupling to hardware and lack of software engineering technologies, the control software in digitally controlled power conversion systems is difficult to design and maintain. This is a natural consequence of a topology- or application-driven design approach. This research work proposes a distributed, modular, open control architecture for power conversion systems to reduce control design complexity, encapsulate and localize design dependencies, reduce unnecessary redesign effort and improve software quality. Dataflow style is chosen as the architectural style for the proposed control architecture based on comparative analysis. The detailed implementation of the dataflow architecture is presented. The resulting dataflow control software is evaluated in comparison to the legacy approach to control design used in industry and academia. The dataflow control software for a 3-phase voltage source inverter is also tested on a real PEBB-based converter system. To further explore the flexibility of control composition that is brought by the dataflow approach, the feasibility of dynamic control reconfiguration is also presented as an important future research direction. / Ph. D.

While increasing number of vehicles on urban roadways create uncontrolled congestion, connectivity among vehicles, traffic lights and other road-side units provide abundant data that paves avenues for novel smart traffic control mechanisms to mitigate traffic congestion and delays. However, increasingly complex vehicular applications have outpaced the computational capabilities of on-board processing units, therefore requiring novel offloading schemes onto additional resources located by the road-side. Adding connectivity and other computational resources on legacy traffic infrastructure may also introduce security vulnerabilities. To ensure that the timeliness and resource constraints of the vehicles using the roadways as well as the applications being deployed on the traffic infrastructure are met, the transportation systems needs to be more predictable. This dissertation discusses three areas that focus on improving the predictability and performance of the connected traffic infrastructure. Firstly, a holistic traffic control strategy is presented that ensures predictable traffic flow by minimizing traffic delays, accounting for unexpected traffic conditions and ensuring timely emergency vehicle traversal through an urban road network. Secondly, a vehicular edge resource management strategy is discussed that incorporates connected traffic lights data to meet timeliness requirements of the vehicular applications. Finally, security vulnerabilities in existing traffic controllers are studied and countermeasures are provided to ensure predictable traffic flow while thwarting attacks on the traffic infrastructure. / Doctor of Philosophy / Exponentially increasing vehicles especially in urban areas create pollution, delays and uncontrolled traffic congestion. However, improved traffic infrastructure brings connectivity among the vehicles, traffic lights, road-side detectors and other equipment, which can be leveraged to design new and advanced traffic control techniques. The initial work in this dissertation provides a traffic control technique that (i) reduces traffic wait times for the vehicles in urban areas, (ii) ensures safe and quick movements of emergency vehicles even through crowded areas, and (iii) ensures that the traffic keeps moving even under unexpected lane closures or roadblocks. As technology advances, connected vehicles are becoming increasingly automated. This allows the car manufacturers to design novel in-vehicle features where the passengers can now stream media-rich content, play augmented reality (AR)-based games and/or get high definition information about the surroundings on their car's display, while the car is driven through the urban traffic. This is made possible by providing additional computing resources along the road-side that the vehicles can utilize wirelessly to ensure passenger's comfort and improved experience of in-vehicle features. In this dissertation, a technique is provided to manage the computational resources which will allow vehicles (and its passengers) to use multiple features simultaneously. As the traffic infrastructure becomes increasingly inter-connected, it also allows malicious actors to exploit vulnerabilities such as modifying traffic lights, interfering with road-side sensors, etc. This can lead to increased traffic wait times and eventually bring down the traffic network. In the final work, one such vulnerability in traffic infrastructure is studied and mitigating measures are provided so that the traffic keeps moving even when an attack is detected. In all, this dissertation aims to improve safety, security and overall experience of the drivers, passengers and the pedestrians using the connected traffic infrastructure.

This thesis presents the design and evaluation of an operating system kernel specially designed for dataflow software. Dataflow is a style of software architecture that is well suited for control and "signal flow" applications. This architecture involves many small processes and lots of inter-process communication, which impose too much overhead on traditional RTOSes. This thesis describes design and implementation of the Dataflow Architecture Real-time Kernel (DARK). DARK is a reconfigurable, multithreaded and preemptive operating system kernel that introduces a special data-driven scheduling strategy for dataflow applications. It uses the underlying hardware for high-speed context switching between the kernel and applications, which is five times faster than the ordinary context switch. The features of the kernel can be configured according to performance requirements without change to the applications. Along with the performance evaluation of DARK, the performance comparison results of DARK with two commercial RTOSes: MicroC/OS-II and Analog Devices VDK++ are also provided. / Master of Science

The structure of an RS232 monitor intended for use in online system maintenance of real-time, industrial process control systems is presented. The monitor is written in PLM86 for use in an Intel 8086 based system in which system software is organized into a set of foreground interrupts and an infinite background loop. The monitor, which resides in the background loop, utilizes a programming technique referred to as psuedo-concurrent programming to eliminate the lockout problem associated with the background loop structure. The psuedo-concurrent technique is explained. Maintenance procedures are described and possible implementations utilizing the monitor are suggested. / Master of Science

This thesis aims to address the real-time planning and nonlinear control of quadrupedal locomotion such that the resulting gaits are robust to various kinds of disturbances. Specifically, this work addresses two scenarios. Namely, a quasi-static formulation in which an inertial appendage (i.e., a tail) is used to assist the quadruped in negating external push disturbances, and an agile formulation which is derived in a manner such that an appendage could easily be added in future work to examine the affect of tails on agile and high-speed motions. Initially, this work presents a unified method in which bio-inspired articulated serpentine robotic tails may be integrated with walking robots, specifically quadrupeds, in order to produce stable and highly robust locomotion. The design and analysis of a holonomically constrained 2 degree of freedom (DOF) tail is shown and its accompanying nonlinear dynamic model is presented. The model created is used to develop a hierarchical control scheme which consists of a high-level path planner and a full-order nonlinear low-level controller. The high-level controller is based on model predictive control (MPC) and acts on a linear inverted pendulum (LIP) model which has been extended to include the forces produced by the tail by augmenting the LIP model with linearized tail dynamics. The MPC is used to generate center of mass (COM) and tail trajectories and is subject to the net ground reaction forces of the system, tail shape, and torque saturation of the tail in order to ensure overall feasibility of locomotion. At the lower level, a full-order nonlinear controller is implemented to track the generated trajectories using quadratic program (QP) based input-output (I-O) feedback linearization which acts on virtual constraints. The analytical results of the proposed approach are verified numerically through simulations using a full-order nonlinear model for the quadrupedal robot, Vision60, augmented with a tail, totaling at 20 DOF. The simulations include a variety of disturbances to show the robustness of the presented hierarchical control scheme. The aforementioned control scheme is then extended in the latter portion of this thesis to achieve more dynamic, agile, and robust locomotion. In particular, we examine the use of a single rigid body model as the template model for the real-time high-level MPC, which is linearized using variational based linearization (VBL) and is solved at 200 Hz as opposed to an event-based manner. The previously defined virtual constraints controller is also extended so as to include a control Lyapunov function (CLF) which contributes to both numerical stability of the QP and aids in stability of the output dynamics. This new hierarchical scheme is validated on the A1 robot, with a total of 18 DOF, through extensive simulations to display agility and robustness to ground height variations and external disturbances. The low-level controller is then further validated through a series of experiments displaying the ability for this algorithm to be readily transferred to hardware platforms. / Master of Science / This thesis aims to address the real-time planning and nonlinear control of four legged walking robots such that the resulting gaits are robust to various kinds of disturbances. Initially, this work presents a method in which a robotic tail can be integrated with legged robots to produce very stable walking patterns. A model is subsequently created to develop a multi-layer control scheme which consists of a high-level path planner, based on a reduced-order model and model predictive control techniques, that determines the trajectory for the quadruped and tail, followed by a low-level controller that considers the full-order dynamics of the robot and tail for robust tracking of the planned trajectory. The reduced-order model considered here enforces quasi-static motions which are slow but generally stable. This formulation is validated numerically through extensive full-order simulations of the Vision60 robot. This work then proceeds to develop an agile formulation using a similar multi-layer structure, but uses a reduced-order model which is more amenable to dynamic walking patterns. The low-level controller is also augmented slightly to provide additional robustness and theoretical guarantees. The latter control algorithm is extensively numerically validated in simulation using the A1 robot to show the large increase in robustness compared to the quasi-static formulation. Finally, this work presents experimental validation of the low-level controller formulated in the latter half of this work.